YOLOv5

New YOLOv5 release

In this release, we provide 2 new architectures of YOLOv5:

• YOLOv5m
• YOLOv5l
  The result is shown as below:

Model	size (pixels)	mAPval 0.5:0.95	mAPval 0.5	mAPtest 0.5:0.95	mAPtest 0.5	Speed V100 (ms)		params (M)
YOLOv5s	640	65.4	93.1	65.4	93.2	6.3		7.3
YOLOv5m	640	66.5	93.9	66.7	93.7	7.9		21.4
YOLOv5l	640	65.8	93.9	66.9	93.8	12.1		47.0

New pre-trained YOLOv3-tiny, YOLOv5s

Re-train YOLOv3-tiny and YOLOv5s

Note: the previous version was trained on incorrect data (Sr, that's my fault)

We present you latest train on new data version with the training result as shown below:

Model	size (pixels)	mAPval 0.5:0.95	mAPval 0.5	mAPtest 0.5:0.95	mAPtest 0.5	Speed V100 (ms)		params (M)
YOLOv5s	640	65.4	93.1	65.4	93.2	6.3		7.3
YOLOv3-tiny	640	55.7	87.9	55.6	87.8	3.4		8.8

Pre-trained weights ver1.1

Pre-trained models version 1.1

In this release, we introduce a list of pre-trained models for academic usages.
Including:

• YOLOv5s
• YOLOv5m
• YOLOv5l
• YOLOv3-tiny

With training experiment shown below:

Model	size (pixels)	mAPval 0.5:0.95	mAPtest 0.5:0.95	mAPval 0.5	Speed V100 (ms)		params (M)
[YOLOv5s][assets_5]	640	59.6	59.6	89.6	4.5		7.3
[YOLOv5m][assets_5]	640	63.7	73.9	91.9	8.5		21.4
[YOLOv5l][assets_5]	640	64.6	65.0	92.1	11.8		47.0
[YOLOv3-tiny][assets_3]	640	49.8	49.7	84.0	3.6		8.8